Iranian Journal of Finance

Abstract In classical risk management models, the assumption of normally distributed
returns leads to an underestimation of the true probability of extreme losses.
Extreme Value Theory (EVT), by focusing on tail behavior, provides an
effective framework for modeling heavy-tailed risks; however, determining the
threshold beyond which tail behavior begins remains one of its main
challenges. In this study, the Good–Usual–Bad (GUB) method introduced by
(González-Sánchez, 2021) and (González-Sánchez & Nave Pineda (2023) is
employed to separate returns into central and extreme components, enabling
the automatic identification of the tail threshold. Subsequently, Value-at-Risk
(VaR) is estimated using the Hill estimator and the Generalized Pareto
2
Iranian Journal of Finance, 2026, Vol. 10, No. 2 (Ebadinezhad, N.)
Distribution (GPD). The dataset consists of daily returns for four asset classes
in the Iranian market over the period 2015-2025: 18-carat gold, the Tehran
Stock Exchange Overall Index (TEDPIX), the USD/IRR exchange rate, and
Bitcoin in Iranian Rial. Empirical results indicate that, at the univariate level,
the GUB method provides accurate estimates for heavy-tailed assets; however,
at the portfolio level, its sensitivity to dependence structures makes it
conservative and prone to overestimating risk. In contrast, dependence-based
models such as the t-student copula yield more reliable portfolio VaR
estimates, and hybrid approaches combining GUB with parametric models and
the modified GUB-based procedures proposed in this study produce more
balanced and efficient risk assessments.

Abstract This study examines the impact of stock market development on key corporate financial outcomes among firms listed on the Tehran Stock Exchange, with a comparative analysis across two distinct capital market periods. The study tests five hypotheses using panel data from 140 listed companies over the period 2011–2024. To capture structural changes in the market, the sample period is divided into two sub-periods: 2011–2017 (first period) and 2018–2024 (second period). The empirical analysis is conducted using panel regression models with firm fixed effects. The results indicate that stock market development has a stronger and more pronounced impact during the second period (2018–2024). Specifically, investment efficiency and corporate financial health improve more substantially in the second period compared to the first. In addition, the cost of capital and reliance on debt financing decline more significantly in the second period. The findings also show a structural shift in corporate investment behavior: under greater stock market development, firms tend to move toward financial investment strategies and away from non-financial asset investment in the later period. This study contributes to the literature by providing comparative evidence on how structural changes and transparency-enhancing reforms in the stock market influence corporate financial outcomes over time. By distinguishing between two development phases of the capital market, the study offers new insights into the effectiveness of market reforms and their implications for corporate financial behavior.

Abstract Accurately predicting stock market movements remains a complex engineering challenge due to inherent non-linearity and volatility. This study addresses this problem by developing a novel artificial intelligence framework that fuses deep learning-based natural language processing with technical analysis for financial forecasting. Its primary engineering application is predicting the daily directional movement of the Tehran Stock Exchange total index, a critical task for automated trading systems and risk management. The core AI contribution is a hybrid architecture that generates contextual text embeddings from Persian news using a Bidirectional Encoder Representations from Transformers (BERT) model and integrates them with technical indicators. These features are processed by a soft-voting ensemble of powerful gradient-boosting algorithms, namely eXtreme Gradient Boosting (XG-Boost) and Light Gradient Boosting Machine (LightGBM). Rigorously validated through an extensive walk-forward testing procedure across 331 temporal windows, the model achieved impressive performance metrics: 84.0% accuracy, 86.1% F1 Score, 87.0% precision, and 85.2% recall. These notable results demonstrate the significant value of domain-specific natural language processing in non-English financial contexts. Furthermore, this work establishes gradient boosting ensembles as a highly efficient, high-performance alternative to Long Short-Term Memory (LSTM) networks for financial forecasting. The proposed methodology demonstrates robust predictive capability for directional movement forecasting in emerging markets.

Abstract The increasing importance of financial risk measurement, driven by regulatory requirements and market uncertainties, necessitates precise forecasting tools for Value at Risk (VaR) and Expected Shortfall (ES). This study introduces three novel models based on Stateful Recurrent Neural Networks (RNNs) and Feedforward Neural Networks (FNNs) for predicting VaR and ES. These models are evaluated against traditional econometric models, including Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models, using a dataset of the Tehran Exchange Divedend and Price Index (TEDPIX) spanning more than three decades. This research presents a comprehensive comparison between Stateful and Non-Stateful RNN models, demonstrating that Stateful RNNs significantly enhance forecasting performance. First, the data is standardized before being fed into machine learning models and then used as input for the models. The results indicate strong predictive performance, while a reduction in the FZ0 loss score highlights the superior accuracy of these models. In the empirical study, the proposed models are applied to the Tehran Exchange Divedend and Price Index (TEDPIX). The models’ performance is assessed using the FZ0 loss function and backtesting methods, such as DQ regression tests and DES tests. The findings reveal that Stateful RNN-based models outperform traditional econometric models in terms of both accuracy and stability. In particular, the STRNN-II and STRNN-III models consistently rank highest, demonstrating their superior ability to capture the complex dynamics of financial risk. Moreover, these models do not rely on strong distributional assumptions, offering greater flexibility than traditional parametric models like GARCH.

Abstract Predicting the movement of financial resources into and out of corporate banking clients is a critical challenge for liquidity risk management and strategic treasury planning. Traditional forecasting models rely predominantly on linear statistical techniques and fail to capture the complex, dynamic interactions between firms and their banking relationships. This study introduces a novel predictive framework grounded in Neural Potential Field Networks (NPFNs). This hybrid architecture integrates artificial neural networks with the theoretical principles of potential field theory drawn from physics and robotics. Formally, each corporate client i at time t is represented as a point in a financial state space, and the model learns a scalar potential function whose negative gradient defines an attraction-repulsion vector field over that state space; a feedforward neural network parameterizes this potential function, and the gradient driving each client's predicted fund flow is obtained by automatic differentiation of the learned potential with respect to the client's state vector, conceptually analogous to artificial potential field methods in robotic motion planning. The proposed model conceptualizes corporate clients as dynamic agents operating within a financial potential field landscape, wherein fund inflows represent attraction forces and outflows represent repulsion forces acting upon liquidity reservoirs. By adapting the gradient-based mechanics of potential fields to model directional liquidity flows between enterprises, the framework captures both short-term transactional volatility and long-term structural patterns in corporate fund behavior. The model produces monthly fund flow forecasts at the individual client-account level over a 12-month out-of-sample evaluation window. The model is trained and validated on a panel dataset comprising 12 large Iranian commercial banks, covering the period 2018 to 2023 and encompassing over 3,400 corporate client accounts. Empirical results demonstrate that the NPFN model achieves a Mean Absolute Percentage Error (MAPE) of 4.73% for inflow prediction and 5.21% for outflow prediction, outperforming LSTM, GRU, and traditional ARIMA benchmarks by statistically significant margins, with all models evaluated on an identical held-out 12-month test period (January-December 2023), an identical feature set, and hyperparameters tuned via grid search on a validation subsample drawn from the preceding 12 months of the training window. Additionally, the model identifies key macroeconomic and microeconomic drivers, including interbank interest rate spreads, corporate leverage ratios, supply chain interconnectedness, and central bank regulatory signals, as primary determinants of potential field gradients. These findings provide actionable intelligence for bank asset-liability management (ALM) committees and offer a theoretically grounded, computationally tractable framework for next-generation corporate liquidity forecasting.

Abstract This study investigates the causal relationship between uncertainty shocks and factor premia in the US equity market. While traditional asset pricing models rely on correlation-based factor structures, this research argues that identifying true alpha necessitates a causal framework. Employing a novel combination of Double Machine Learning (DML) and Causal Forest algorithms on Kenneth French's five-factor data and VIX-based uncertainty shocks from 2001 to 2025, robust evidence is provided demonstrating that uncertainty shocks exert significant causal effects on factor returns. The findings reveal that the market factor (MKT-RF) and the size factor (SMB) exhibit strong negative causal responses to uncertainty shocks. In contrast, the profitability factor (RMW) demonstrates a positive and significant effect. The value (HML) and investment (CMA) factors, however, show no statistically detectable causal effect. Substantial heterogeneity in these effects is documented across high- and low-volatility regimes, with the market's negative response being substantially larger in high-uncertainty periods. From an economic significance perspective, a long-only trading strategy based on causal signals delivers a competitive Sharpe ratio of 0.437, with a substantially lower maximum drawdown (-32.5%) compared to the market portfolio (-51.4%). The results survive rigorous robustness checks and statistical tests, providing evidence consistent with a causal effect of uncertainty shocks on factor premia. This study contributes to the growing literature on causal machine learning in finance and offers practical implications for risk management and portfolio construction.